The effect of context-dependent information and sentence constructions on perceived humanness of an agent in a Turing test